Eiffel: Efficient and Fair Scheduling in Adaptive Federated Learning

Emerging machine learning (ML) technologies, in combination with the increasing computational power of mobile devices, lead to the extensive adoption of ML-based applications. Different from conventional model training that needs to collect all the user data in centralized cloud servers, federated learning (FL) has recently drawn increasing research attention as it enables privacy-preserving model training. With FL, decentralized edge devices in participation, train their model copies locally over their siloed datasets, and periodically synchronize the model parameters. However, model training is computationally extensive which easily drains the battery of mobile devices. In addition, due to the uneven distribution of siloed datasets, the shared model may become biased. To address the efficiency and fairness concerns in a resource-constrained federated learning setting, in this paper, we propose Eiffel to judiciously select mobile devices to participate in the global model aggregation, and adaptively adjust the frequency of local and global model updates. Eiffel aims to make scheduling and coordination for the federated learning towards both resource efficiency and model fairness. We have conducted theoretical analysis of Eiffel from the perspectives of fairness and convergence. Extensive experiments with a wide variety of real-world datasets and models, both on a networked prototype system and in a larger-scale simulated environment, have demonstrated that while maintaining similar accuracy performance, Eiffel outperforms existing baselines with respect to reducing communication overhead by up to 6× for higher efficiency and improving the fairness metric by up to 57% compared to the state-of-the-art algorithms.

1 Introduction

With the ever-growing computation capability and the extensive adoption of mobile devices (e.g., smartphones, wearable medical devices, sensory equipment) in today's era of Internet-of-Things, an astronomical amount of data are generated daily over the network. According to a recent survey of Cisco, IoT devices will account for 50% (14.7 billion) of all global networked devices by 2023 [1]. Each edge device is producing massive amount of data every year, which can be naturally leveraged by user-interactive applications driven by machine learning techniques. Typically, a machine learning model is trained in a centralized fashion where a datacenter gathers input data from all the participating edge devices. As one might anticipate, this is not a suitable method of model training for edge devices due to privacy sensitivity of user data and communication burden incurred by transferring massive raw data.